Videos
Generate video and audio with a job queue
Submit a job, check its status, download the finished MP4. Built for models that take a minute or two per clip and produce the highest quality.
4–15 s
Clip length per job
768p
Max output, self-hosted MiniMax-H3
50–150 s
Typical full-quality generation
What the offline API is
Offline generation is queue-and-poll. You describe the clip you want, we put the job in a queue, a GPU worker claims it, and some time later a finished MP4 is ready to download. You never hold a connection open for the duration of the generation, and a slow minute costs you nothing but waiting.
- Submit a job with
POST /v1/videos. You get an id back immediately. - Check its status with
GET /v1/videos/{id}until it iscompleted. - Download the result from
GET /v1/videos/{id}/content. - One request, video and audio. MiniMax-H3 returns synchronized stereo audio in the same MP4.
- Same
tf_API key and error envelope as the rest of the API.
What you can ask for
| Task | Input | Use it for |
|---|---|---|
t2va | A text prompt | Text to video and audio. |
fl2va | A first and/or last frame, plus optional text | Animate an image, or bridge two frames. |
ref2va | Reference images, video, or audio, plus optional text | Keep a character, setting, or voice consistent. |
Offline or realtime?
| Offline video | Realtime video | |
|---|---|---|
| Shape | Submit, poll, download a file | Open a session, watch frames, send commands live |
| Latency | Minutes per clip | Frames as they are generated |
| Quality | Large models, full-clip generation | Small streaming models |
| Connection | Plain HTTPS | HTTPS plus WebRTC |
| Status | Specified | Preview |
Pick offline when you want a finished clip and can wait. Pick realtime when a person is watching and steering.